I have been working my way through the paper titled "DeepMPC: Learning Deep Latent Features for Model Predictive Control", and I had a conceptual question. Is the time varying non linear dynamics model described in this paper simply a LSTM?
I have been digging deeper into LSTM's and it seems like that is what it is, but they only say that they "propose a novel deep architecture which is able to model dynamics conditioned on learned latent properties and a multi-stage pre-training algorithm that avoids common problems in training recurrent neural networks". This sounds an awful lot like an LSMT. If this isn't an LSTM, what is it?
Here is a link to the paper for reference: http://deepmpc.cs.cornell.edu/DeepMPC.pdf
The 'deep predictive model' is detailed in section IV and illustrated in figure 5.
Also, I apologize if this is the wrong place to ask this. If it is, could you please direct me to the correct stack exchange community?